Re: [PATCH net-next v7 1/7] net: introduce rstreason to detect why the RST is sent
On Tue, Apr 23, 2024 at 7:57 PM Simon Horman wrote:
>
> On Tue, Apr 23, 2024 at 10:17:31AM +0800, Jason Xing wrote:
> > On Tue, Apr 23, 2024 at 10:14 AM Jason Xing
> > wrote:
> > >
> > > Hi Simon,
> > >
> > > On Tue, Apr 23, 2024 at 2:28 AM Simon Horman wrote:
> > > >
> > > > On Mon, Apr 22, 2024 at 11:01:03AM +0800, Jason Xing wrote:
> > > >
> > > > ...
> > > >
> > > > > diff --git a/include/net/rstreason.h b/include/net/rstreason.h
> > > >
> > > > ...
> > > >
> > > > > +/**
> > > > > + * There are three parts in order:
> > > > > + * 1) reset reason in MPTCP: only for MPTCP use
> > > > > + * 2) skb drop reason: relying on drop reasons for such as passive
> > > > > reset
> > > > > + * 3) independent reset reason: such as active reset reasons
> > > > > + */
> > > >
> > > > Hi Jason,
> > > >
> > > > A minor nit from my side.
> > > >
> > > > '/**' denotes the beginning of a Kernel doc,
> > > > but other than that, this comment is not a Kernel doc.
> > > >
> > > > FWIIW, I would suggest providing a proper Kernel doc for enum
> > > > sk_rst_reason.
> > > > But another option would be to simply make this a normal comment,
> > > > starting with "/* There are"
> > >
> > > Thanks Simon. I'm trying to use the kdoc way to make it right :)
> > >
> > > How about this one:
> > > /**
> > > * enum sk_rst_reason - the reasons of socket reset
> > > *
> > > * The reason of skb drop, which is used in DCCP/TCP/MPTCP protocols.
> >
> > s/skb drop/sk reset/
> >
> > Sorry, I cannot withdraw my previous email in time.
> >
> > > *
> > > * There are three parts in order:
> > > * 1) skb drop reasons: relying on drop reasons for such as passive
> > > reset
> > > * 2) independent reset reasons: such as active reset reasons
> > > * 3) reset reasons in MPTCP: only for MPTCP use
> > > */
> > > ?
> > >
> > > I chose to mimic what enum skb_drop_reason does in the
> > > include/net/dropreason-core.h file.
> > >
> > > > +enum sk_rst_reason {
> > > > + /**
> > > > +* Copy from include/uapi/linux/mptcp.h.
> > > > +* These reset fields will not be changed since they adhere to
> > > > +* RFC 8684. So do not touch them. I'm going to list each
> > > > definition
> > > > +* of them respectively.
> > > > +*/
> > >
> > > Thanks to you, I found another similar point where I smell something
> > > wrong as in the above code. I'm going to replace '/**' with '/*' since
> > > it's only a comment, not a kdoc.
>
> Likewise, thanks Jason.
>
> I haven't had time to look at v8 properly,
> but I see that kernel-doc is happy with the changed
> you have made there as discussed above.
>
Thank you, Simon. I learned something new about the coding style.
Besides, some other nit problems have been spotted by Matt. I will fix
them if it's required to send a new version.
Re: [PATCH net-next v7 1/7] net: introduce rstreason to detect why the RST is sent
On Tue, Apr 23, 2024 at 10:17:31AM +0800, Jason Xing wrote:
> On Tue, Apr 23, 2024 at 10:14 AM Jason Xing wrote:
> >
> > Hi Simon,
> >
> > On Tue, Apr 23, 2024 at 2:28 AM Simon Horman wrote:
> > >
> > > On Mon, Apr 22, 2024 at 11:01:03AM +0800, Jason Xing wrote:
> > >
> > > ...
> > >
> > > > diff --git a/include/net/rstreason.h b/include/net/rstreason.h
> > >
> > > ...
> > >
> > > > +/**
> > > > + * There are three parts in order:
> > > > + * 1) reset reason in MPTCP: only for MPTCP use
> > > > + * 2) skb drop reason: relying on drop reasons for such as passive
> > > > reset
> > > > + * 3) independent reset reason: such as active reset reasons
> > > > + */
> > >
> > > Hi Jason,
> > >
> > > A minor nit from my side.
> > >
> > > '/**' denotes the beginning of a Kernel doc,
> > > but other than that, this comment is not a Kernel doc.
> > >
> > > FWIIW, I would suggest providing a proper Kernel doc for enum
> > > sk_rst_reason.
> > > But another option would be to simply make this a normal comment,
> > > starting with "/* There are"
> >
> > Thanks Simon. I'm trying to use the kdoc way to make it right :)
> >
> > How about this one:
> > /**
> > * enum sk_rst_reason - the reasons of socket reset
> > *
> > * The reason of skb drop, which is used in DCCP/TCP/MPTCP protocols.
>
> s/skb drop/sk reset/
>
> Sorry, I cannot withdraw my previous email in time.
>
> > *
> > * There are three parts in order:
> > * 1) skb drop reasons: relying on drop reasons for such as passive
> > reset
> > * 2) independent reset reasons: such as active reset reasons
> > * 3) reset reasons in MPTCP: only for MPTCP use
> > */
> > ?
> >
> > I chose to mimic what enum skb_drop_reason does in the
> > include/net/dropreason-core.h file.
> >
> > > +enum sk_rst_reason {
> > > + /**
> > > +* Copy from include/uapi/linux/mptcp.h.
> > > +* These reset fields will not be changed since they adhere to
> > > +* RFC 8684. So do not touch them. I'm going to list each
> > > definition
> > > +* of them respectively.
> > > +*/
> >
> > Thanks to you, I found another similar point where I smell something
> > wrong as in the above code. I'm going to replace '/**' with '/*' since
> > it's only a comment, not a kdoc.
Likewise, thanks Jason.
I haven't had time to look at v8 properly,
but I see that kernel-doc is happy with the changed
you have made there as discussed above.
Re: [PATCH net-next v7 1/7] net: introduce rstreason to detect why the RST is sent
On Tue, Apr 23, 2024 at 10:14 AM Jason Xing wrote:
>
> Hi Simon,
>
> On Tue, Apr 23, 2024 at 2:28 AM Simon Horman wrote:
> >
> > On Mon, Apr 22, 2024 at 11:01:03AM +0800, Jason Xing wrote:
> >
> > ...
> >
> > > diff --git a/include/net/rstreason.h b/include/net/rstreason.h
> >
> > ...
> >
> > > +/**
> > > + * There are three parts in order:
> > > + * 1) reset reason in MPTCP: only for MPTCP use
> > > + * 2) skb drop reason: relying on drop reasons for such as passive reset
> > > + * 3) independent reset reason: such as active reset reasons
> > > + */
> >
> > Hi Jason,
> >
> > A minor nit from my side.
> >
> > '/**' denotes the beginning of a Kernel doc,
> > but other than that, this comment is not a Kernel doc.
> >
> > FWIIW, I would suggest providing a proper Kernel doc for enum sk_rst_reason.
> > But another option would be to simply make this a normal comment,
> > starting with "/* There are"
>
> Thanks Simon. I'm trying to use the kdoc way to make it right :)
>
> How about this one:
> /**
> * enum sk_rst_reason - the reasons of socket reset
> *
> * The reason of skb drop, which is used in DCCP/TCP/MPTCP protocols.
s/skb drop/sk reset/
Sorry, I cannot withdraw my previous email in time.
> *
> * There are three parts in order:
> * 1) skb drop reasons: relying on drop reasons for such as passive
> reset
> * 2) independent reset reasons: such as active reset reasons
> * 3) reset reasons in MPTCP: only for MPTCP use
> */
> ?
>
> I chose to mimic what enum skb_drop_reason does in the
> include/net/dropreason-core.h file.
>
> > +enum sk_rst_reason {
> > + /**
> > +* Copy from include/uapi/linux/mptcp.h.
> > +* These reset fields will not be changed since they adhere to
> > +* RFC 8684. So do not touch them. I'm going to list each definition
> > +* of them respectively.
> > +*/
>
> Thanks to you, I found another similar point where I smell something
> wrong as in the above code. I'm going to replace '/**' with '/*' since
> it's only a comment, not a kdoc.
>
> Thanks,
> Jason
Re: [PATCH net-next v7 1/7] net: introduce rstreason to detect why the RST is sent
Hi Simon,
On Tue, Apr 23, 2024 at 2:28 AM Simon Horman wrote:
>
> On Mon, Apr 22, 2024 at 11:01:03AM +0800, Jason Xing wrote:
>
> ...
>
> > diff --git a/include/net/rstreason.h b/include/net/rstreason.h
>
> ...
>
> > +/**
> > + * There are three parts in order:
> > + * 1) reset reason in MPTCP: only for MPTCP use
> > + * 2) skb drop reason: relying on drop reasons for such as passive reset
> > + * 3) independent reset reason: such as active reset reasons
> > + */
>
> Hi Jason,
>
> A minor nit from my side.
>
> '/**' denotes the beginning of a Kernel doc,
> but other than that, this comment is not a Kernel doc.
>
> FWIIW, I would suggest providing a proper Kernel doc for enum sk_rst_reason.
> But another option would be to simply make this a normal comment,
> starting with "/* There are"
Thanks Simon. I'm trying to use the kdoc way to make it right :)
How about this one:
/**
* enum sk_rst_reason - the reasons of socket reset
*
* The reason of skb drop, which is used in DCCP/TCP/MPTCP protocols.
*
* There are three parts in order:
* 1) skb drop reasons: relying on drop reasons for such as passive
reset
* 2) independent reset reasons: such as active reset reasons
* 3) reset reasons in MPTCP: only for MPTCP use
*/
?
I chose to mimic what enum skb_drop_reason does in the
include/net/dropreason-core.h file.
> +enum sk_rst_reason {
> + /**
> +* Copy from include/uapi/linux/mptcp.h.
> +* These reset fields will not be changed since they adhere to
> +* RFC 8684. So do not touch them. I'm going to list each definition
> +* of them respectively.
> +*/
Thanks to you, I found another similar point where I smell something
wrong as in the above code. I'm going to replace '/**' with '/*' since
it's only a comment, not a kdoc.
Thanks,
Jason
Re: [PATCH net-next v7 1/7] net: introduce rstreason to detect why the RST is sent
On Mon, Apr 22, 2024 at 11:01:03AM +0800, Jason Xing wrote:
...
> diff --git a/include/net/rstreason.h b/include/net/rstreason.h
...
> +/**
> + * There are three parts in order:
> + * 1) reset reason in MPTCP: only for MPTCP use
> + * 2) skb drop reason: relying on drop reasons for such as passive reset
> + * 3) independent reset reason: such as active reset reasons
> + */
Hi Jason,
A minor nit from my side.
'/**' denotes the beginning of a Kernel doc,
but other than that, this comment is not a Kernel doc.
FWIIW, I would suggest providing a proper Kernel doc for enum sk_rst_reason.
But another option would be to simply make this a normal comment,
starting with "/* There are"
> +enum sk_rst_reason {
...
Re: [PATCH net-next v7 1/7] net: introduce rstreason to detect why the RST is sent
On 22/04/2024 11:17, Jason Xing wrote:> On Mon, Apr 22, 2024 at 4:47 PM Matthieu Baerts wrote: >> On 22/04/2024 05:01, Jason Xing wrote: >>> From: Jason Xing (...) >>> diff --git a/include/net/rstreason.h b/include/net/rstreason.h >>> new file mode 100644 >>> index ..c57bc5413c17 >>> --- /dev/null >>> +++ b/include/net/rstreason.h >>> @@ -0,0 +1,144 @@ >>> +/* SPDX-License-Identifier: GPL-2.0-or-later */ >>> + >>> +#ifndef _LINUX_RSTREASON_H >>> +#define _LINUX_RSTREASON_H >>> +#include >>> +#include >>> + >>> +#define DEFINE_RST_REASON(FN, FNe) \ >>> + FN(MPTCP_RST_EUNSPEC) \ >>> + FN(MPTCP_RST_EMPTCP)\ >>> + FN(MPTCP_RST_ERESOURCE) \ >>> + FN(MPTCP_RST_EPROHIBIT) \ >>> + FN(MPTCP_RST_EWQ2BIG) \ >>> + FN(MPTCP_RST_EBADPERF) \ >>> + FN(MPTCP_RST_EMIDDLEBOX)\ >> >> Small detail: should it not make more sense to put the ones linked to >> MPTCP at the end? I mean I guess MPTCP should be treated in second >> priority: CONFIG_MPTCP could not be set, and the ones linked to TCP >> should be more frequent, etc. > > Do you mean that I need to adjust the order: 1) tcp reasons first, 2) > independent reasons, 3) mptcp reasons ? Correct, it looks like it is a more "natural" order. > Reasonable. I will do it :) Thanks! Cheers, Matt -- Sponsored by the NGI0 Core fund.
Re: [PATCH net-next v7 1/7] net: introduce rstreason to detect why the RST is sent
Hello Matthieu,
On Mon, Apr 22, 2024 at 4:47 PM Matthieu Baerts wrote:
>
> Hi Jason,
>
> On 22/04/2024 05:01, Jason Xing wrote:
> > From: Jason Xing
> >
> > Add a new standalone file for the easy future extension to support
> > both active reset and passive reset in the TCP/DCCP/MPTCP protocols.
>
> Thank you for looking at that!
Thanks for the review!
>
> (...)
>
> > diff --git a/include/net/rstreason.h b/include/net/rstreason.h
> > new file mode 100644
> > index ..c57bc5413c17
> > --- /dev/null
> > +++ b/include/net/rstreason.h
> > @@ -0,0 +1,144 @@
> > +/* SPDX-License-Identifier: GPL-2.0-or-later */
> > +
> > +#ifndef _LINUX_RSTREASON_H
> > +#define _LINUX_RSTREASON_H
> > +#include
> > +#include
> > +
> > +#define DEFINE_RST_REASON(FN, FNe) \
> > + FN(MPTCP_RST_EUNSPEC) \
> > + FN(MPTCP_RST_EMPTCP)\
> > + FN(MPTCP_RST_ERESOURCE) \
> > + FN(MPTCP_RST_EPROHIBIT) \
> > + FN(MPTCP_RST_EWQ2BIG) \
> > + FN(MPTCP_RST_EBADPERF) \
> > + FN(MPTCP_RST_EMIDDLEBOX)\
>
> Small detail: should it not make more sense to put the ones linked to
> MPTCP at the end? I mean I guess MPTCP should be treated in second
> priority: CONFIG_MPTCP could not be set, and the ones linked to TCP
> should be more frequent, etc.
Do you mean that I need to adjust the order: 1) tcp reasons first, 2)
independent reasons, 3) mptcp reasons ?
Reasonable. I will do it :)
>
> > + FN(NOT_SPECIFIED) \
> > + FN(NO_SOCKET) \
> > + FNe(MAX)
>
> (...)
>
> > +/* Convert reset reasons in MPTCP to our own enum type */
> > +static inline enum sk_rst_reason convert_mptcpreason(u32 reason)
> > +{
> > + switch (reason) {
> > + case MPTCP_RST_EUNSPEC:
> > + return SK_RST_REASON_MPTCP_RST_EUNSPEC;
> > + case MPTCP_RST_EMPTCP:
> > + return SK_RST_REASON_MPTCP_RST_EMPTCP;
> > + case MPTCP_RST_ERESOURCE:
> > + return SK_RST_REASON_MPTCP_RST_ERESOURCE;
> > + case MPTCP_RST_EPROHIBIT:
> > + return SK_RST_REASON_MPTCP_RST_EPROHIBIT;
> > + case MPTCP_RST_EWQ2BIG:
> > + return SK_RST_REASON_MPTCP_RST_EWQ2BIG;
> > + case MPTCP_RST_EBADPERF:
> > + return SK_RST_REASON_MPTCP_RST_EBADPERF;
> > + case MPTCP_RST_EMIDDLEBOX:
> > + return SK_RST_REASON_MPTCP_RST_EMIDDLEBOX;
> > + default:
> > + /**
> > + * It should not happen, or else errors may occur
> > + * in MPTCP layer
> > + */
> > + return SK_RST_REASON_ERROR;
> > + }
> > +}
>
> If this helper is only used on MPTCP, maybe better to move it to
> net/mptcp/protocol.h (and to patch 5/7?)? We tried to isolate MPTCP code.
Roger that. I will move the helper into protocol.h as well as the patch itself.
>
> Also, maybe it is just me, but I'm not a big fan of the helper name:
> convert_mptcpreason() (same for the "drop" one). I think it should at
> least mention its "origin" (rst reason): e.g. something like
> (sk_)rst_reason_convert_mptcp or (sk_)rst_convert_mptcp_reason() (or
> mptcp_to_rst_reason())?
>
> And (sk_)rst_reason_convert_(skb_)drop() (or skb_drop_to_rst_reason())?
I agree with you. Actually I had a local patch where I used
sk_rst_reason_skbdrop() and sk_rst_reason_mptcpreason().
Interestingly, I changed them in this patch series due to the function
name being too long (which is my initial thought).
I will use sk_rst_convert_xxx_reason() as you suggested.
>
> > +/* Convert reset reasons in MPTCP to our own enum type */
>
> I don't think this part is linked to MPTCP, right?
Ah, copy-paste syndrome... Sorry, I will correct it.
>
> > +static inline enum sk_rst_reason convert_dropreason(enum skb_drop_reason
> > reason)
> > +{
> > + switch (reason) {
> > + case SKB_DROP_REASON_NOT_SPECIFIED:
> > + return SK_RST_REASON_NOT_SPECIFIED;
> > + case SKB_DROP_REASON_NO_SOCKET:
> > + return SK_RST_REASON_NO_SOCKET;
> > + default:
> > + /* If we don't have our own corresponding reason */
> > + return SK_RST_REASON_NOT_SPECIFIED;
> > + }
> > +}
>
> (This helper could be introduced in patch 4/7 because it is not used
> before, but I'm fine either ways.)
Good. It makes more sense.
Thanks,
Jason
Re: [PATCH net-next v7 1/7] net: introduce rstreason to detect why the RST is sent
Hi Jason,
On 22/04/2024 05:01, Jason Xing wrote:
> From: Jason Xing
>
> Add a new standalone file for the easy future extension to support
> both active reset and passive reset in the TCP/DCCP/MPTCP protocols.
Thank you for looking at that!
(...)
> diff --git a/include/net/rstreason.h b/include/net/rstreason.h
> new file mode 100644
> index ..c57bc5413c17
> --- /dev/null
> +++ b/include/net/rstreason.h
> @@ -0,0 +1,144 @@
> +/* SPDX-License-Identifier: GPL-2.0-or-later */
> +
> +#ifndef _LINUX_RSTREASON_H
> +#define _LINUX_RSTREASON_H
> +#include
> +#include
> +
> +#define DEFINE_RST_REASON(FN, FNe) \
> + FN(MPTCP_RST_EUNSPEC) \
> + FN(MPTCP_RST_EMPTCP)\
> + FN(MPTCP_RST_ERESOURCE) \
> + FN(MPTCP_RST_EPROHIBIT) \
> + FN(MPTCP_RST_EWQ2BIG) \
> + FN(MPTCP_RST_EBADPERF) \
> + FN(MPTCP_RST_EMIDDLEBOX)\
Small detail: should it not make more sense to put the ones linked to
MPTCP at the end? I mean I guess MPTCP should be treated in second
priority: CONFIG_MPTCP could not be set, and the ones linked to TCP
should be more frequent, etc.
> + FN(NOT_SPECIFIED) \
> + FN(NO_SOCKET) \
> + FNe(MAX)
(...)
> +/* Convert reset reasons in MPTCP to our own enum type */
> +static inline enum sk_rst_reason convert_mptcpreason(u32 reason)
> +{
> + switch (reason) {
> + case MPTCP_RST_EUNSPEC:
> + return SK_RST_REASON_MPTCP_RST_EUNSPEC;
> + case MPTCP_RST_EMPTCP:
> + return SK_RST_REASON_MPTCP_RST_EMPTCP;
> + case MPTCP_RST_ERESOURCE:
> + return SK_RST_REASON_MPTCP_RST_ERESOURCE;
> + case MPTCP_RST_EPROHIBIT:
> + return SK_RST_REASON_MPTCP_RST_EPROHIBIT;
> + case MPTCP_RST_EWQ2BIG:
> + return SK_RST_REASON_MPTCP_RST_EWQ2BIG;
> + case MPTCP_RST_EBADPERF:
> + return SK_RST_REASON_MPTCP_RST_EBADPERF;
> + case MPTCP_RST_EMIDDLEBOX:
> + return SK_RST_REASON_MPTCP_RST_EMIDDLEBOX;
> + default:
> + /**
> + * It should not happen, or else errors may occur
> + * in MPTCP layer
> + */
> + return SK_RST_REASON_ERROR;
> + }
> +}
If this helper is only used on MPTCP, maybe better to move it to
net/mptcp/protocol.h (and to patch 5/7?)? We tried to isolate MPTCP code.
Also, maybe it is just me, but I'm not a big fan of the helper name:
convert_mptcpreason() (same for the "drop" one). I think it should at
least mention its "origin" (rst reason): e.g. something like
(sk_)rst_reason_convert_mptcp or (sk_)rst_convert_mptcp_reason() (or
mptcp_to_rst_reason())?
And (sk_)rst_reason_convert_(skb_)drop() (or skb_drop_to_rst_reason())?
> +/* Convert reset reasons in MPTCP to our own enum type */
I don't think this part is linked to MPTCP, right?
> +static inline enum sk_rst_reason convert_dropreason(enum skb_drop_reason
> reason)
> +{
> + switch (reason) {
> + case SKB_DROP_REASON_NOT_SPECIFIED:
> + return SK_RST_REASON_NOT_SPECIFIED;
> + case SKB_DROP_REASON_NO_SOCKET:
> + return SK_RST_REASON_NO_SOCKET;
> + default:
> + /* If we don't have our own corresponding reason */
> + return SK_RST_REASON_NOT_SPECIFIED;
> + }
> +}
(This helper could be introduced in patch 4/7 because it is not used
before, but I'm fine either ways.)
> +#endif
Cheers,
Matt
--
Sponsored by the NGI0 Core fund.
